A single phototransistor mimics retinal bipolar cell function while detecting visible and infrared light, enabling low-power image processing through programmable ON and OFF photoresponses.
(Nanowerk Spotlight) A camera mounted on a vehicle may produce a clear image in daylight. That same sensor becomes unreliable at night, in fog, or when facing oncoming headlights. It records light, but not always usefully. Interpreting that light, separating signal from noise, identifying motion, sharpening the boundaries of objects, still depends on downstream processing that takes time and consumes energy. The sensor itself plays no active role in shaping the information it collects.
Human vision works differently. Long before visual signals reach the brain, they are filtered and organized inside the retina. Among the key components are bipolar cells. These neurons receive input from photoreceptors and split it into ON and OFF signals. The ON channel responds to increases in light, while the OFF channel responds to decreases. This early division makes it easier for the brain to recognize edges, detect contrast, and respond to visual changes quickly.
Electronic systems have struggled to replicate this process. Most photodetectors respond to light passively. They capture intensity but do not classify or process it. Some devices can detect infrared as well as visible light, but few can change how they respond to the same optical signal based on context. Even fewer can adjust their output without continuous external power. These limitations restrict how artificial vision systems perform in complex environments such as autonomous driving or mobile robotics.
The problem is not just detection. It is the lack of early-stage interpretation. To match the efficiency of biological vision, a sensor must not only react to light but also reshape the signal as it arrives. This requires materials that can store information between operations, respond across a broad range of wavelengths, and switch behavior based on short, low-power commands.
The researchers designed a device that performs not only light detection but also preprocessing, directly at the sensor level. Their phototransistor uses a layered architecture built on a germanium base. The channel that conducts the electrical signal is made from tungsten diselenide, a two-dimensional semiconductor that interacts strongly with visible light.
Between this channel and the gate lies a thin sheet of hexagonal boron nitride, which serves as an insulating barrier. Above that sits a floating gate made of multilayer graphene. The entire structure rests on a thin silicon dioxide layer grown on top of a germanium substrate. Each layer plays a distinct role in capturing, modulating, and storing information.
Design and characteristics of the floating-gate composite structure phototransistor. a) Simplified schematic of signal transmission process in the human retinal system and working mechanism of ON and OFF bipolar cells in photoresponse. b) Conceptual schematic of FG-CSPT with image sensing and processing processes. The device not only receives external image information, but also pre-processes and modulates signals by programming its photoresponse through gate voltage. c) Schematic of the device structure. d) Optical microscope image of the phototransistor (orange, green and grey outlines for WSe2, hBN, and MLG, respectively). e) Raman spectrum of the device. (Image: Reprinted from DOI:10.1002/advs.202512649, CC BY) (click on image to enlarge)
The key innovation is the use of a floating gate that stores electrical charge non-volatilely. Instead of requiring constant gate voltage to maintain its state, the device responds to short voltage pulses. A positive pulse pushes electrons through the boron nitride layer and into the floating gate, changing the electric field experienced by the tungsten diselenide channel. A negative pulse pushes holes instead. This shifts the channel from one type of charge carrier to another, changing its conductivity and the direction of its response to incoming light.
This programmable switching mimics the behavior of bipolar retinal cells. In the human retina, ON and OFF bipolar cells respond differently to the same light signal. The phototransistor replicates this distinction. When exposed to the same wavelength of light, it produces a positive electrical signal under one gate setting and a negative signal under another. The result is a single device that can separate light input into structured ON and OFF signals, similar to biological vision.
To verify this behavior, the authors exposed the device to green light at 532 nanometers and infrared light at 1550 nanometers. Under green light, the tungsten diselenide layer absorbs photons and generates pairs of electrons and holes. These charges move under the influence of an internal electric field that depends on the gate’s stored charge. When the gate is programmed with a negative pulse, the channel becomes rich in electrons, and the current flows in one direction. A positive pulse reverses the field, and the current flows in the opposite direction.
Under infrared illumination, the mechanism changes. The tungsten diselenide layer does not absorb much infrared light on its own, but the underlying germanium substrate does. Germanium has a narrow bandgap that allows it to absorb lower-energy photons. When infrared light strikes the germanium, it generates electron-hole pairs. Some of these electrons become trapped at the interface between germanium and the silicon dioxide layer. These trapped charges modify the electric field in the channel above, producing a photogating effect. This acts like an additional gate voltage that changes the behavior of the transistor without direct electrical input.
Meanwhile, the graphene layers also absorb some infrared light. Because graphene has no bandgap, it can generate hot carriers that cross the junction into the tungsten diselenide channel. This photo-thermionic effect contributes further to the device’s infrared sensitivity. Together, the photogating effect in germanium and the thermionic emission from graphene allow the device to respond to infrared signals with the same programmable ON and OFF behavior it shows under visible light.
The measurements confirm the design’s effectiveness. The device switches between states with a change in gate pulse of only 30 volts, and holds that state for over 2000 seconds without additional power. The current output varies by a factor of up to one million between ON and OFF states.
Responsivity, which measures the amount of current generated per unit of incoming light, reaches 1.3 amperes per watt under green light in the ON state and negative 2.98 amperes per watt in the OFF state. Under infrared light, the responsivity is lower but still measurable, with values up to 0.57 milliamperes per watt in the ON state and negative 1.08 milliamperes per watt in the OFF state.
To demonstrate practical use, the authors integrated the device into an image processing task. They constructed a basic convolution kernel, a common tool used in computer vision to enhance edges and sharpen images. The phototransistor replaces both the sensor and the processing unit. It collects the image data and shapes the signal directly, based on its programmable state. This allows for faster processing and lower energy use, since there is no need to transfer data to a separate processor for initial filtering.
In one test, the team simulated vehicle imaging under both daylight and nighttime conditions. Under normal light, the device behaves like a visible light camera. Under low light or fog, the infrared sensitivity allows it to continue operating. Using its programmable ON and OFF responses, it can highlight changes in brightness and detect contours in real time. The same device, without changing hardware, performs sensing and contrast enhancement by switching its state through short electrical pulses.
Compared to earlier attempts at retinal emulation or floating-gate phototransistors, this design combines multiple essential features in a single platform. It responds to visible and infrared light. It switches between ON and OFF responses to the same optical input. It stores state without constant energy input. And it performs early-stage signal shaping without external computation. These features move artificial vision closer to the functional efficiency of biological systems.
This work demonstrates a phototransistor that detects both visible and infrared light while reproducing the signal-separating function of bipolar retinal cells. It combines non-volatile electrical control with broad-spectrum optical response in a single structure, allowing the same device to sense and preprocess images with programmable polarity. By integrating memory, detection, and simple processing in one unit, it reduces the need for external computation and lowers power consumption. The approach provides a compact and adaptable foundation for machine vision systems that must operate reliably in variable light conditions and respond in real time.
For authors and communications departmentsclick to open
Lay summary
Prefilled posts
Nanowerk Newsletter
Get our Nanotechnology Spotlight updates to your inbox!
Thank you!
You have successfully joined our subscriber list.
Become a Spotlight guest author! Join our large and growing group of guest contributors. Have you just published a scientific paper or have other exciting developments to share with the nanotechnology community? Here is how to publish on nanowerk.com.
